Skip to content

Pulse

Overview

Pulse is Palm's AI intelligence layer — a natural language interface that lets treasury teams interact with their cash data through conversation. It encompasses the Palm Chat (customer-facing), Palm MCP (internal), AI Digest (homepage), and future agentic capabilities like scheduled report workflows.

The core insight: treasury teams spend hours asking questions that an LLM with access to the right data can answer in seconds. Pulse bridges the gap between raw data in BigQuery and the insights treasurers need for daily decision-making.

Key architectural decisions: schema-first workflow (eliminates LLM hallucination on table names), infrastructure-level security via GCP service account impersonation (prompt-level security is insufficient), and structured JSON responses validated against real records before rendering.

For detailed architecture and terminology, see fundamentals.md

Top Jobs & Desired Outcomes

Full history: jobs.md

⚡ 1. Deliver proactive treasury insights without requiring user-initiated queries

Desired Outcomes: - Minimize the delay between an actionable insight existing in data and the user becoming aware of it - Reduce dependency on users knowing the right questions to ask - Increase the frequency of valuable, unsolicited recommendations reaching treasury teams

"We need to be proactive. We need to show the user something worth seeing." — Emma (2026-03-10)

Source: Palm Internal (2026-03-10) — Emerging (validated by CS experience with ON's categorization analysis workflow)

Key Pain Points

Full history: pain-points.md

  • LLM hallucination on table/column names — Without schema-first enforcement, LLMs invent plausible table names (Source: Palm Internal 2026-03-09)
  • Prompt-level security is insufficient — System prompts can be bypassed; data isolation must be infrastructure-enforced (Source: Palm Internal 2026-03-09)
  • No response evaluation framework — No systematic way to check correctness or detect hallucinations post-chat (Source: Palm Internal 2026-03-09)
  • No observability for MCP usage — Once MCP access is given, Palm has no visibility into queries or response quality (Source: Palm Internal 2026-03-09)
  • Users see AI as "just another chatbot" — Don't understand it can be a proactive assistant; need templates to guide them (Source: Palm Internal 2026-03-10)
  • Reactive model limits value delivery — Highest-value insights are the ones users don't know they need (Source: Palm Internal 2026-03-10)

Key Opportunities

  • Personalized digest — Learn from user behavior to customize homepage blocks per role (e.g., head of treasury vs cash manager)
  • Learning loop — Use chat logs to discover repeated questions and surface as digest blocks or proactive alerts
  • Natural language scenarios — Create forecast scenarios through conversation instead of button-based UI
  • Scheduled agent workflows — Recurring analysis that runs automatically and delivers PDF/Excel reports
  • Flexible file ingestion — LLM-assisted upload for new data types (budgets, investment terms) without building static UIs
  • Unstructured document context — Policy documents, investment policies as retrieval context for Pulse (requires RAG)

Open Questions

  • [ ] What questions do treasury teams most commonly ask that an AI assistant could answer?
  • [ ] What proactive alerts would be most valuable for daily treasury operations?
  • [ ] How do teams feel about AI-generated insights vs manual analysis?
  • [ ] Do different user roles (head of treasury vs cash manager) need fundamentally different digest content?
  • [ ] Would users trust AI-generated scenarios enough to act on them without manual verification?
  • [ ] What's the right accuracy disclaimer / beta labeling for customer-facing AI responses?
  • [ ] How should we evaluate correctness systematically — per-query evals, sampling, user feedback?
  • [ ] What's the cost trajectory as usage scales, and what efficiency strategies should we pursue?

Last updated: 2026-03-10 | Sources: 2 transcripts (view all)